[Arm-netbook] EOMA server standard

[luke.leighton]

On Fri, Oct 26, 2012 at 5:36 PM, Gordan Bobic <gordan at bobich.net> wrote:
> On 10/26/2012 05:13 PM, luke.leighton wrote:
>> On Fri, Oct 26, 2012 at 4:39 PM, Gordan Bobic<gordan at bobich.net>  wrote:
>>
>>> Hang on - 24 pins out of 68 are reserved for RGB?!
>>
>>   28.  leaving 40 for everything else, including 8 for GND and 5V.
>> scarey, huh? :)
>>
>>> If there was ever a
>>> good reason to use a composite video output - this is it.
>>
>> ah.  considered that, 18 months ago.  rejected it for 2 reasons.  3.
>>
>> a) composite video however isn't a lowest-common-denominator across a
>> wide range of SoCs
>
> So add a chip to do it, maybe?

 no.  absolutely not.  whatever the cost of that chip is, it's
automatically too much.  see b), below, which i note conspicuously you
didn't reply to or say "fair enough" or anything :)

 it could be $0.50 for that chip, but automatically that's too much,
esp. as that could be *greater* than the entire profit margin for a
mass-volume 100 million+ units product.

>> b) for a 5 to 7in tablet with a low-cost 800x480 LCD (or a 480x320 or
>> even less), in the critical price-sensitive bracket, you'd need a
>> converter IC from composite video over to the low-cost RGB/TTL 800x480
>> LCDs.
>>
>> c) for the hi-res range, i don't believe that composite video can
>> really cope with 2048x2048 at 30fps (OMAP3530) or even 1920x1080p60, can
>> it?
>
> Not 100% sure. VGA is 15-pins (5 wires), of which most aren't used, and
> that is good for up to at least 2048x1536.

 ah.  i know why. power.  VGA, being 75 ohms, consumes really quite
large amounts of current to drive those lines.  and it's serial rather
than parallel, and a square law applies.


> Perhaps this may be of use:
> http://en.wikipedia.org/wiki/Serial_digital_interface
>
>> plus, i hate to think of the cost of ICs for conversion from
>> composite video over to dual-channel LVDS.
>
> On the scale of the educated guess of a retail price of an EOMA card,
> how much difference is this _really_ going to make, compared to how much
> it cripples the standard?

 as it doesn't cripple the standard, this is a non-issue.

>> so... no.
>>
>> plus, jammy buggers that we are, those 40^H^H32 remaining pins are
>> *just* enough for USB3, 1000Eth, SATA and I2C, with 12 pins to spare!
>
> Fair enough. But if you had those other 20-odd pins, you could perhaps
> avoid having to have multiple standards.

 well, no, you wouldn't, because, again, due to the "nothing must be
optional" rule, anything that was optional (as already discussed and
eliminated) *or* "multiplexed" which is another no-no would have to be
forcibly available as part of the standard.  *all* SoCs would be
forced to provide *both* of the multiplex functions.

 let's say you multiplexed PCIe x2 onto those 28 pins: you'd now
*completely* eliminate all the low-cost SoCs out there that don't have
PCIe x2 from the standard.

 for this reason it's actually better to have completely
non-compatible connectors, even.  yes i can see that there's a case
either way.

l.